queen_of_embedded_egy_20k

masri-embed-student-10k

نموذج تعلّم تمثيلات (embeddings) للعامية المصرية مبني على:

• Base model: sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2
• Fine-tuned on a subset (~20000 عينة) من داتاسيت EgyTriplets-250K
• Training objective: triplet loss على (anchor, positive, negative) للجمل المصرية

Usage (Python)

from transformers import AutoTokenizer, AutoModel
import torch
import torch.nn.functional as F

model_id = "Ahmedhisham/queen_of_embedded_egy_20k"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModel.from_pretrained(model_id)

def encode(texts, max_length=128, device="cuda" if torch.cuda.is_available() else "cpu"):
    model.to(device)
    model.eval()
    with torch.no_grad():
        enc = tokenizer(
            texts,
            padding=True,
            truncation=True,
            max_length=max_length,
            return_tensors="pt",
        ).to(device)
        out = model(**enc)
        last_hidden = out.last_hidden_state
        mask = enc["attention_mask"].unsqueeze(-1).expand(last_hidden.size()).float()
        masked = last_hidden * mask
        summed = masked.sum(dim=1)
        counts = mask.sum(dim=1).clamp(min=1e-9)
        emb = summed / counts
        emb = F.normalize(emb, p=2, dim=-1)
    return emb

# Example
texts = [
    "عايز أروح الساحل أغير جو وأرتاح شوية",
    "محتاج أجازة على البحر كام يوم",
    "بحب أقرأ كتب عن الذكاء الاصطناعي",
]

embs = encode(texts)
sim = torch.matmul(embs, embs.T)
print(sim)